MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202611107123 A) filed by Kulvinder Singh; Sanjeev Dhawan; and Priyanka Sharma on September 06, 2026, for A Firefly-Based Multi-Objective Feature Selection System For Robust Text Sentiment Classification.
Inventors include Kulvinder Singh; Sanjeev Dhawan; and Priyanka Sharma.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: The present invention provides a robust multi-objective feature selection and sentiment classification system and method for high-dimensional textual data which is robust with respect to the data dimensionality and size. Textual records are labeled and then preprocessed by removing noise, lowercasing, eliminating URLs, punctuation, removing stop words, tokenizing, stemming and lemmatizing. Two or more feature representations are created, such as Term Frequency-Inverse Document Frequency (TF-IDF), Word2Vec, GloVe and BERT. A Multi-Objective Firefly Optimization module contains subsets of candidate features encoded as binary feature- selection vectors and measures the performance of the candidate sets via the use of a plurality of classifiers that are: Naïve Bayes, Random Forest and Multi-Layer Perceptron (MLP). Ensemble-guided fitness value computed as a weighted classification accuracy, recall and normalized feature-subset penalty. To search for an optimized feature subset, the firefly movement algorithm, and the sigmoid based binary conversion and mutation are repeated. An adaptive pruning mechanism based on distribution of feature selection is introduced and dynamically changes the threshold to retain features. Various strategies on class-imbalance such as Smote, cost sensitive learning and cost-sensitive loss function are used. A simplified subset of the features is then used for sentiment classification and assessed using accuracy, precision, recall, Fmeasure, number of features and convergence stability. The disclosed system offers a multi-objective and classifier independent feature-selection method that provides a high-level idea of which features to use for good sentiment-classification performance while reducing the feature dimensionality.
Disclaimer: Curated by HT Syndication.
